Metabolomics seeks to achieve a comprehensive quantitative analysis of the wide range of metabolites in biological samples. Analytical chemistry, and specifically 1H NMR, can provide spectra detailing the vast array of physio-chemical properties of the sample. The identification of characteristics in the spectra is known as metabolomic fingerprinting. However, the thousands of variables in the 1H NMR spectra, .relating to hundreds of metabolites, make the data sets extremely complex and difficult to interpret. The aim of this research was to develop analytical techniques to classify biological samples, which are able to identify the metabolites responsible for the differences between classes. Novel methods based on genetic programming have been developed which provide models that are able to classify samples at least as well as currently used chemometric pattern recognition approaches. Moreover these techniques have the advantage that they are significantly easier to interpret in terms of the original spectrum. Preprocessing the spectra using wavelet transforms has further increased the interpretability of the models. This adaptive binning method integrates the data in such a way that the bins represent peaks in the original spectra thereby reducing the dimensionality whilst maintaining the information. The preprocessing and genetic programming methods have been combined and developed in response to additio~al problems faced by time resolved metabolomics data, where both intra-individual (time dependent) and inter-individual variation are present. The methods were employed in the analysis of TSE infected sheep and cattle and, in both cases, enabled a diagnostic of the disease to be determined, allowing the specific compounds responsible for the disease to be identified.
Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:490269 |
Date | January 2007 |
Creators | Davis, Richard |
Publisher | University of York |
Source Sets | Ethos UK |
Detected Language | English |
Type | Electronic Thesis or Dissertation |
Page generated in 0.0019 seconds